Loading...

The Effect of Travel Time Uncertainty on the Results of Discrete Route Choice Models

Khavarian, Khashayar | 2024

0 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 57349 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Kermanshah, Mohammad; Samimi, Amir; Pourzahedi, Hossein
  7. Abstract:
  8. In this research, the effect of uncertainty on the Logit and probit discrete choice models’ results during a route choice situation is assessed. Researchers measure average travel time or calculate it with traffic assignment modules. However, just like any other random variable, the estimated or measured average travel time does not have a certain value and has measurement errors. Logit and Probit models are the most commonly used discrete choice models yet none of them are capable of considering the independent variable’s uncertainty. The amount of difference between models overlooking independent variables’ value uncertainty and the variation of the model capable of considering these uncertainties during estimation process is not investigated in any study. One obstacle towards this comparison is the researcher’s unawareness of independent’s variable uncertainty and its distribution. In the first step, a model is estimated based on observed routes’ travel time and network characteristics to estimate the travel time standard variation. Next, a discrete choice model based on logit and probit structure capable of including variable’s value uncertainty in the estimation phase is used. This model is called “model with uncertain value for independent variable” throughout the text and is considered the basis of comparison with simple logit and probit models. The comparison shows that simple model coefficients are 7 to 14 percent different than those of model with uncertain value for independent variable. In the next step, the sensitivity of obtained result to uncertainty distribution and the uncertainty distribution throughout sample population is assessed. At last it is observed that four times difference between coefficients and elasticity values of models are possible with the travel time standard deviation for each observation of the sample population equal and between 0.05 to 0.5. Based on previous research, these boundary values of uncertainty are probable for travel time. Therefore, the obtained results show the importance of uncertainty inclusion in the model estimation. We also show that Probit model with uncertain value for independent variable is more sensitive to uncertainty values that the logit counterpart. The pattern of change in model’s outcomes with respect to uncertainty values are similar for both probit and logit models and with uncertainty incensement, the route choice probabilities’ elasticity regarding travel time increases and reaches a maximum value then it starts to decrease reaching an almost constant value or the simple model’s elasticity. The details of described change pattern is related to model type (logit or probit) and the uncertainty distribution type (Normal, Log-normal, Gamma, and Weibull). The model with uncertain value for independent variable’s application is depicted using a toy network and a real case based on City of Lyon, France. Here, we emphasized the importance of travel time uncertainty inclusion in model estimation phase and its required assumption’s effect on the results are presented. Moreover, the presented method for travel time uncertainty estimation and the models with uncertain value for independent variables can be used for route choice modeling in future studies
  9. Keywords:
  10. Route Choice Modeling ; Sensitivity Analysis ; Logit Model ; Probit Model ; Path Finding ; Travel Time Uncertainty ; Travel Time Variability

 Digital Object List

 Bookmark